/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
* So far the only truly scary part is that Linux relies on the inode cache
* to be up to date. Don't you dare break a callback and expect an fstat
* to give you meaningful information. This appears to be fixed in the 2.1
- * development kernels. As it is we can fix this now by intercepting the
+ * development kernels. As it is we can fix this now by intercepting the
* stat calls.
*/
}
#endif
-extern int BlobScan(struct dcache * afile, afs_int32 ablob);
+extern int BlobScan(struct dcache * afile, afs_int32 ablob, afs_int32 *ablobOut);
/* This is a complete rewrite of afs_readdir, since we can make use of
* filldir instead of afs_readdir_move. Note that changes to vcache/dcache
struct dcache *tdc;
int code;
int offset;
- int dirpos;
+ afs_int32 dirpos;
struct DirEntry *de;
struct DirBuffer entry;
ino_t ino;
tdc = afs_GetDCache(avc, (afs_size_t) 0, treq, &origOffset, &tlen, 1);
len = tlen;
if (!tdc) {
- code = -ENOENT;
+ code = -EIO;
goto out;
}
ObtainWriteLock(&avc->lock, 811);
offset = (int) fp->f_pos;
#endif
while (1) {
- dirpos = BlobScan(tdc, offset);
- if (!dirpos)
+ code = BlobScan(tdc, offset, &dirpos);
+ if (code || !dirpos)
break;
code = afs_dir_GetVerifiedBlob(tdc, dirpos, &entry);
UpgradeSToWLock(&avc->lock, 814);
avc->f.states |= CCorrupt;
}
- code = -ENOENT;
+ code = -EIO;
goto unlock_out;
}
if ((avc->f.states & CForeign) == 0 && (ntohl(de->fid.vnode) & 1)) {
type = DT_DIR;
} else if ((tvc = afs_FindVCache(&afid, 0, 0))) {
- if (tvc->mvstat) {
+ if (tvc->mvstat != AFS_MVSTAT_FILE) {
type = DT_DIR;
} else if (((tvc->f.states) & (CStatd | CTruth))) {
/* CTruth will be set if the object has
/* clean up from afs_FindVCache */
afs_PutVCache(tvc);
}
- /*
+ /*
* If this is NFS readdirplus, then the filler is going to
* call getattr on this inode, which will deadlock if we're
* holding the GLOCK.
struct vcache *vcp = VTOAFS(FILE_INODE(fp));
cred_t *credp = crref();
struct AFS_FLOCK flock;
-
+
/* Convert to a lock format afs_lockctl understands. */
memset(&flock, 0, sizeof(flock));
flock.l_type = flp->fl_type;
#endif /* F_GETLK64 && F_GETLK != F_GETLK64 */
AFS_GLOCK();
- if ((vcp->f.states & CRO)) {
- if (flp->fl_type == F_WRLCK) {
- code = EBADF;
- } else {
- code = 0;
- }
- AFS_GUNLOCK();
- crfree(credp);
- return code;
- }
code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
AFS_GUNLOCK();
- if ((code == 0 || flp->fl_type == F_UNLCK) &&
+ if ((code == 0 || flp->fl_type == F_UNLCK) &&
(cmd == F_SETLK || cmd == F_SETLKW)) {
code = afs_posix_lock_file(fp, flp);
if (code && flp->fl_type != F_UNLCK) {
return 0;
}
}
-
+
/* Convert flock back to Linux's file_lock */
flp->fl_type = flock.l_type;
flp->fl_pid = flock.l_pid;
code = afs_convert_code(afs_lockctl(vcp, &flock, cmd, credp));
AFS_GUNLOCK();
- if ((code == 0 || flp->fl_type == F_UNLCK) &&
+ if ((code == 0 || flp->fl_type == F_UNLCK) &&
(cmd == F_SETLK || cmd == F_SETLKW)) {
flp->fl_flags &=~ FL_SLEEP;
code = flock_lock_file_wait(fp, flp);
#ifdef STRUCT_FILE_OPERATIONS_HAS_READ_ITER
.read_iter = afs_linux_read_iter,
.write_iter = afs_linux_write_iter,
+# if !defined(HAVE_LINUX___VFS_READ)
.read = new_sync_read,
.write = new_sync_write,
+# endif
#elif defined(HAVE_LINUX_GENERIC_FILE_AIO_READ)
.aio_read = afs_linux_aio_read,
.aio_write = afs_linux_aio_write,
#if defined(STRUCT_FILE_OPERATIONS_HAS_SENDFILE)
.sendfile = generic_file_sendfile,
#endif
-#if defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE)
+#if defined(STRUCT_FILE_OPERATIONS_HAS_SPLICE) && !defined(HAVE_LINUX_DEFAULT_FILE_SPLICE_READ)
# if defined(HAVE_LINUX_ITER_FILE_SPLICE_WRITE)
.splice_write = iter_file_splice_write,
# else
* AFS Linux dentry operations
**********************************************************************/
-/* fix_bad_parent() : called if this dentry's vcache is a root vcache
- * that has its mvid (parent dir's fid) pointer set to the wrong directory
- * due to being mounted in multiple points at once. fix_bad_parent()
- * calls afs_lookup() to correct the vcache's mvid, as well as the volume's
- * dotdotfid and mtpoint fid members.
- * Parameters:
- * dp - dentry to be checked.
- * credp - credentials
- * vcp, pvc - item's and parent's vcache pointer
- * Return Values:
- * None.
- * Sideeffects:
- * This dentry's vcache's mvid will be set to the correct parent directory's
- * fid.
- * This root vnode's volume will have its dotdotfid and mtpoint fids set
- * to the correct parent and mountpoint fids.
- */
-
-static inline void
-fix_bad_parent(struct dentry *dp, cred_t *credp, struct vcache *vcp, struct vcache *pvc)
-{
- struct vcache *avc = NULL;
-
- /* force a lookup, so vcp->mvid is fixed up */
- afs_lookup(pvc, (char *)dp->d_name.name, &avc, credp);
- if (!avc || vcp != avc) { /* bad, very bad.. */
- afs_Trace4(afs_iclSetp, CM_TRACE_TMP_1S3L, ICL_TYPE_STRING,
- "check_bad_parent: bad pointer returned from afs_lookup origvc newvc dentry",
- ICL_TYPE_POINTER, vcp, ICL_TYPE_POINTER, avc,
- ICL_TYPE_POINTER, dp);
- }
- if (avc)
- AFS_RELE(AFSTOV(avc));
-
- return;
-}
-
/* afs_linux_revalidate
* Ensure vcache is stat'd before use. Return 0 if entry is valid.
*/
cred_t *credp;
int code;
- if (afs_shuttingdown)
+ if (afs_shuttingdown != AFS_RUNNING)
return EIO;
AFS_GLOCK();
goto out;
}
-#ifdef notyet
- /* Make this a fast path (no crref), since it's called so often. */
- if (vcp->states & CStatd) {
- struct vcache *pvc = VTOAFS(dp->d_parent->d_inode);
-
- if (*dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
- if (vcp->mvid->Fid.Volume != pvc->fid.Fid.Volume) { /* bad parent */
- credp = crref();
- AFS_GLOCK();
- fix_bad_parent(dp); /* check and correct mvid */
- AFS_GUNLOCK();
- crfree(credp);
- }
- }
- afs_DestroyAttr(vattr);
- return 0;
- }
-#endif
-
/* This avoids the crref when we don't have to do it. Watch for
* changes in afs_getattr that don't get replicated here!
*/
if (vcp->f.states & CStatd &&
- (!afs_fakestat_enable || vcp->mvstat != 1) &&
+ (!afs_fakestat_enable || vcp->mvstat != AFS_MVSTAT_MTPT) &&
!afs_nfsexporter &&
(vType(vcp) == VDIR || vType(vcp) == VLNK)) {
code = afs_CopyOutAttrs(vcp, vattr);
}
static afs_uint32
-parent_vcache_dv(struct inode *inode, cred_t *credp)
+parent_vcache_dv(struct inode *inode, cred_t *credp, int locked)
{
int free_cred = 0;
struct vcache *pvcp;
* us. The fake entry is the one with the useful DataVersion.
*/
pvcp = VTOAFS(inode);
- if (pvcp->mvstat == 1 && afs_fakestat_enable) {
+ if (pvcp->mvstat == AFS_MVSTAT_MTPT && afs_fakestat_enable) {
struct vrequest treq;
struct afs_fakestat_state fakestate;
+ if (!locked) {
+ AFS_GLOCK();
+ }
if (!credp) {
credp = crref();
free_cred = 1;
if (free_cred)
crfree(credp);
afs_PutFakeStat(&fakestate);
+ if (!locked) {
+ AFS_GUNLOCK();
+ }
}
return hgetlo(pvcp->f.m.DataVersion);
}
/* Validate a dentry. Return 1 if unchanged, 0 if VFS layer should re-evaluate.
* In kernels 2.2.10 and above, we are passed an additional flags var which
* may have either the LOOKUP_FOLLOW OR LOOKUP_DIRECTORY set in which case
- * we are advised to follow the entry if it is a link or to make sure that
+ * we are advised to follow the entry if it is a link or to make sure that
* it is a directory. But since the kernel itself checks these possibilities
* later on, we shouldn't have to do it until later. Perhaps in the future..
*
parent = dget_parent(dp);
pvcp = VTOAFS(parent->d_inode);
- if ((vcp->mvstat == 1) || (vcp->mvstat == 2) ||
- (pvcp->mvstat == 1 && afs_fakestat_enable)) { /* need to lock */
+ if ((vcp->mvstat != AFS_MVSTAT_FILE) ||
+ (pvcp->mvstat == AFS_MVSTAT_MTPT && afs_fakestat_enable)) { /* need to lock */
credp = crref();
AFS_GLOCK();
locked = 1;
}
- if (locked && vcp->mvstat == 1) { /* mount point */
- if (vcp->mvid && (vcp->f.states & CMValid)) {
- int tryEvalOnly = 0;
- int code = 0;
- struct vrequest *treq = NULL;
-
- code = afs_CreateReq(&treq, credp);
- if (code) {
- dput(parent);
- goto bad_dentry;
- }
- if ((strcmp(dp->d_name.name, ".directory") == 0)) {
- tryEvalOnly = 1;
- }
- if (tryEvalOnly)
- code = afs_TryEvalFakeStat(&vcp, &fakestate, treq);
- else
- code = afs_EvalFakeStat(&vcp, &fakestate, treq);
- afs_DestroyReq(treq);
- if ((tryEvalOnly && vcp->mvstat == 1) || code) {
- /* a mount point, not yet replaced by its directory */
- dput(parent);
- goto bad_dentry;
- }
- }
- } else
- if (locked && *dp->d_name.name != '/' && vcp->mvstat == 2) { /* root vnode */
- if (vcp->mvid->Fid.Volume != pvcp->f.fid.Fid.Volume) { /* bad parent */
- fix_bad_parent(dp, credp, vcp, pvcp); /* check and correct mvid */
+ if (locked) {
+ if (vcp->mvstat == AFS_MVSTAT_MTPT) {
+ if (vcp->mvid.target_root && (vcp->f.states & CMValid)) {
+ int tryEvalOnly = 0;
+ int code = 0;
+ struct vrequest *treq = NULL;
+
+ code = afs_CreateReq(&treq, credp);
+ if (code) {
+ dput(parent);
+ goto bad_dentry;
+ }
+ if ((strcmp(dp->d_name.name, ".directory") == 0)) {
+ tryEvalOnly = 1;
+ }
+ if (tryEvalOnly)
+ code = afs_TryEvalFakeStat(&vcp, &fakestate, treq);
+ else
+ code = afs_EvalFakeStat(&vcp, &fakestate, treq);
+ afs_DestroyReq(treq);
+ if ((tryEvalOnly && vcp->mvstat == AFS_MVSTAT_MTPT) || code) {
+ /* a mount point, not yet replaced by its directory */
+ dput(parent);
+ goto bad_dentry;
+ }
}
+ } else if (vcp->mvstat == AFS_MVSTAT_ROOT && *dp->d_name.name != '/') {
+ osi_Assert(vcp->mvid.parent != NULL);
}
+ }
#ifdef notdef
/* If the last looker changes, we should make sure the current
}
#endif
- parent_dv = parent_vcache_dv(parent->d_inode, credp);
+ parent_dv = parent_vcache_dv(parent->d_inode, credp, locked);
/* If the parent's DataVersion has changed or the vnode
* is longer valid, we need to do a full lookup. VerifyVCache
if (locked && (parent_dv > dp->d_time || !(vcp->f.states & CStatd))) {
struct vattr *vattr = NULL;
int code;
+ int lookup_good;
code = afs_lookup(pvcp, (char *)dp->d_name.name, &tvc, credp);
- if (!tvc || tvc != vcp) {
+
+ if (code) {
+ /* We couldn't perform the lookup, so we're not okay. */
+ lookup_good = 0;
+
+ } else if (tvc == vcp) {
+ /* We got back the same vcache, so we're good. */
+ lookup_good = 1;
+
+ } else if (tvc == VTOAFS(dp->d_inode)) {
+ /* We got back the same vcache, so we're good. This is
+ * different from the above case, because sometimes 'vcp' is
+ * not the same as the vcache for dp->d_inode, if 'vcp' was a
+ * mtpt and we evaluated it to a root dir. In rare cases,
+ * afs_lookup might not evalute the mtpt when we do, or vice
+ * versa, so the previous case will not succeed. But this is
+ * still 'correct', so make sure not to mark the dentry as
+ * invalid; it still points to the same thing! */
+ lookup_good = 1;
+
+ } else {
+ /* We got back a different file, so we're definitely not
+ * okay. */
+ lookup_good = 0;
+ }
+
+ if (!lookup_good) {
dput(parent);
/* Force unhash; the name doesn't point to this file
* anymore. */
/* unlikely--the vcache entry hasn't changed */
dput(parent);
+
} else {
-#ifdef notyet
- /* If this code is ever enabled, we should use dget_parent to handle
- * getting the parent, and dput() to dispose of it. See above for an
- * example ... */
- pvcp = VTOAFS(dp->d_parent->d_inode);
- if (hgetlo(pvcp->f.m.DataVersion) > dp->d_time)
- goto bad_dentry;
-#endif
- /* No change in parent's DataVersion so this negative
- * lookup is still valid. BUT, if a server is down a
- * negative lookup can result so there should be a
- * liftime as well. For now, always expire.
- */
+ /* 'dp' represents a cached negative lookup. */
- goto bad_dentry;
+ parent = dget_parent(dp);
+ pvcp = VTOAFS(parent->d_inode);
+ parent_dv = parent_vcache_dv(parent->d_inode, credp, locked);
+
+ if (parent_dv > dp->d_time || !(pvcp->f.states & CStatd)
+ || afs_IsDynroot(pvcp)) {
+ dput(parent);
+ goto bad_dentry;
+ }
+
+ dput(parent);
}
good_dentry:
bad_dentry:
if (have_submounts(dp))
valid = 1;
- else
+ else
valid = 0;
goto done;
}
{
struct dentry *target;
- /* avoid symlink resolution limits when resolving; we cannot contribute to
- * an infinite symlink loop */
+ /*
+ * Avoid symlink resolution limits when resolving; we cannot contribute to
+ * an infinite symlink loop.
+ *
+ * On newer kernels the field has moved to the private nameidata structure
+ * so we can't adjust it here. This may cause ELOOP when using a path with
+ * 40 or more directories that are not already in the dentry cache.
+ */
+#if defined(STRUCT_TASK_STRUCT_HAS_TOTAL_LINK_COUNT)
current->total_link_count--;
+#endif
target = canonical_dentry(path->dentry->d_inode);
#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
dp->d_op = &afs_dentry_operations;
#endif
- dp->d_time = parent_vcache_dv(dip, credp);
+ dp->d_time = parent_vcache_dv(dip, credp, 1);
d_instantiate(dp, ip);
}
AFS_GLOCK();
code = afs_lookup(VTOAFS(dip), (char *)comp, &vcp, credp);
-
- if (vcp) {
+
+ if (!code) {
struct vattr *vattr = NULL;
struct vcache *parent_vc = VTOAFS(dip);
#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
dp->d_op = &afs_dentry_operations;
#endif
- dp->d_time = parent_vcache_dv(dip, credp);
+ dp->d_time = parent_vcache_dv(dip, credp, 1);
AFS_GUNLOCK();
d_prune_aliases(ip);
#ifdef STRUCT_DENTRY_OPERATIONS_HAS_D_AUTOMOUNT
- ip->i_flags |= S_AUTOMOUNT;
+ /* Only needed if this is a volume root */
+ if (vcp->mvstat == 2)
+ ip->i_flags |= S_AUTOMOUNT;
#endif
}
+ /*
+ * Take an extra reference so the inode doesn't go away if
+ * d_splice_alias drops our reference on error.
+ */
+ if (ip)
+#ifdef HAVE_LINUX_IHOLD
+ ihold(ip);
+#else
+ igrab(ip);
+#endif
+
newdp = d_splice_alias(ip, dp);
done:
/* It's ok for the file to not be found. That's noted by the caller by
* seeing that the dp->d_inode field is NULL.
*/
- if (!code || code == ENOENT)
- return newdp;
- else
+ if (!code || code == ENOENT) {
+ /*
+ * d_splice_alias can return an error (EIO) if there is an existing
+ * connected directory alias for this dentry.
+ */
+ if (!IS_ERR(newdp)) {
+ iput(ip);
+ return newdp;
+ } else {
+ d_add(dp, ip);
+ /*
+ * Depending on the kernel version, d_splice_alias may or may
+ * not drop the inode reference on error. If it didn't, do it
+ * here.
+ */
+#if defined(D_SPLICE_ALIAS_LEAK_ON_ERROR)
+ iput(ip);
+#endif
+ return NULL;
+ }
+ } else {
+ if (ip)
+ iput(ip);
return ERR_PTR(afs_convert_code(code));
+ }
}
static int
VTOAFS(dir), (char *)__dp->d_name.name,
credp);
if (!code) {
- tvc->mvid = (void *) __name;
+ tvc->mvid.silly_name = __name;
crhold(credp);
if (tvc->uncred) {
crfree(tvc->uncred);
}
-/* afs_linux_ireadlink
+/* afs_linux_ireadlink
* Internal readlink which can return link contents to user or kernel space.
* Note that the buffer is NOT supposed to be null-terminated.
*/
struct uio tuio;
struct iovec iov;
+ memset(&tuio, 0, sizeof(tuio));
+ memset(&iov, 0, sizeof(iov));
+
setup_uio(&tuio, &iov, target, (afs_offs_t) 0, maxlen, UIO_READ, seg);
code = afs_readlink(VTOAFS(ip), &tuio, credp);
crfree(credp);
}
#if !defined(USABLE_KERNEL_PAGE_SYMLINK_CACHE)
-/* afs_linux_readlink
+/* afs_linux_readlink
* Fill target (which is in user space) with contents of symlink.
*/
static int
/* afs_linux_follow_link
* a file system dependent link following routine.
*/
+#if defined(HAVE_LINUX_INODE_OPERATIONS_FOLLOW_LINK_NO_NAMEIDATA)
+static const char *afs_linux_follow_link(struct dentry *dentry, void **link_data)
+#else
static int afs_linux_follow_link(struct dentry *dentry, struct nameidata *nd)
+#endif
{
int code;
char *name;
name = kmalloc(PATH_MAX, GFP_NOFS);
if (!name) {
+#if defined(HAVE_LINUX_INODE_OPERATIONS_FOLLOW_LINK_NO_NAMEIDATA)
+ return ERR_PTR(-EIO);
+#else
return -EIO;
+#endif
}
AFS_GLOCK();
AFS_GUNLOCK();
if (code < 0) {
+#if defined(HAVE_LINUX_INODE_OPERATIONS_FOLLOW_LINK_NO_NAMEIDATA)
+ return ERR_PTR(code);
+#else
return code;
+#endif
}
name[code] = '\0';
+#if defined(HAVE_LINUX_INODE_OPERATIONS_FOLLOW_LINK_NO_NAMEIDATA)
+ return *link_data = name;
+#else
nd_set_link(nd, name);
return 0;
+#endif
}
+#if defined(HAVE_LINUX_INODE_OPERATIONS_PUT_LINK_NO_NAMEIDATA)
+static void
+afs_linux_put_link(struct inode *inode, void *link_data)
+{
+ char *name = link_data;
+
+ if (name && !IS_ERR(name))
+ kfree(name);
+}
+#else
static void
afs_linux_put_link(struct dentry *dentry, struct nameidata *nd)
{
if (name && !IS_ERR(name))
kfree(name);
}
+#endif /* HAVE_LINUX_INODE_OPERATIONS_PUT_LINK_NO_NAMEIDATA */
#endif /* USABLE_KERNEL_PAGE_SYMLINK_CACHE */
99999); /* not a possible code value */
code = afs_rdwr(avc, auio, UIO_READ, 0, credp);
-
+
afs_Trace4(afs_iclSetp, CM_TRACE_READPAGE, ICL_TYPE_POINTER, ip,
ICL_TYPE_POINTER, pp, ICL_TYPE_INT32, cnt, ICL_TYPE_INT32,
code);
if (tdc) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, credp, treq);
- afs_PutDCache(tdc);
+ afs_PutDCache(tdc);
}
ReleaseWriteLock(&avc->lock);
}
* locked */
static inline int
afs_linux_prepare_writeback(struct vcache *avc) {
- if (avc->f.states & CPageWrite) {
- return AOP_WRITEPAGE_ACTIVATE;
+ pid_t pid;
+ struct pagewriter *pw;
+
+ pid = MyPidxx2Pid(MyPidxx);
+ /* Prevent recursion into the writeback code */
+ spin_lock(&avc->pagewriter_lock);
+ list_for_each_entry(pw, &avc->pagewriters, link) {
+ if (pw->writer == pid) {
+ spin_unlock(&avc->pagewriter_lock);
+ return AOP_WRITEPAGE_ACTIVATE;
+ }
}
- avc->f.states |= CPageWrite;
+ spin_unlock(&avc->pagewriter_lock);
+
+ /* Add ourselves to writer list */
+ pw = osi_Alloc(sizeof(struct pagewriter));
+ pw->writer = pid;
+ spin_lock(&avc->pagewriter_lock);
+ list_add_tail(&pw->link, &avc->pagewriters);
+ spin_unlock(&avc->pagewriter_lock);
+
return 0;
}
static inline void
afs_linux_complete_writeback(struct vcache *avc) {
- avc->f.states &= ~CPageWrite;
+ struct pagewriter *pw, *store;
+ pid_t pid;
+ struct list_head tofree;
+
+ INIT_LIST_HEAD(&tofree);
+ pid = MyPidxx2Pid(MyPidxx);
+ /* Remove ourselves from writer list */
+ spin_lock(&avc->pagewriter_lock);
+ list_for_each_entry_safe(pw, store, &avc->pagewriters, link) {
+ if (pw->writer == pid) {
+ list_del(&pw->link);
+ /* osi_Free may sleep so we need to defer it */
+ list_add_tail(&pw->link, &tofree);
+ }
+ }
+ spin_unlock(&avc->pagewriter_lock);
+ list_for_each_entry_safe(pw, store, &tofree, link) {
+ list_del(&pw->link);
+ osi_Free(pw, sizeof(struct pagewriter));
+ }
}
/* Writeback a given page syncronously. Called with no AFS locks held */
struct iovec iovec;
int f_flags = 0;
+ memset(&tuio, 0, sizeof(tuio));
+ memset(&iovec, 0, sizeof(iovec));
+
buffer = kmap(pp) + offset;
base = page_offset(pp) + offset;
void
afs_fill_inode(struct inode *ip, struct vattr *vattr)
{
-
if (vattr)
vattr2inode(ip, vattr);
+#ifdef STRUCT_ADDRESS_SPACE_HAS_BACKING_DEV_INFO
ip->i_mapping->backing_dev_info = afs_backing_dev_info;
+#endif
/* Reset ops if symlink or directory. */
if (S_ISREG(ip->i_mode)) {
ip->i_op = &afs_file_iops;